diff --git a/README.md b/README.md index a2f0524..ac9e59b 100644 --- a/README.md +++ b/README.md @@ -11,7 +11,7 @@ To use the old but cheap Tesla K20Xm with openAI Clip you need to downgrade torc ### CUDA - nVidia Driver NVIDIA-Linux-x86_64-470.82.01.run - `sudo apt install nvidia-settings` -- Build `cuda_check.c` with `nvcc -o check_cuda check_cuda.c -lcuda` +- Build `check_cuda.c` with `nvcc -o check_cuda check_cuda.c -lcuda` - Verify the output of check_cuda: `Compute Capability: 3.5` ## Installing @@ -25,7 +25,9 @@ To use the old but cheap Tesla K20Xm with openAI Clip you need to downgrade torc ### Building PyTorch switch to gcc8 + `sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-8 8` + `sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-8 8` Verify @@ -33,7 +35,7 @@ Verify `conda install astunparse numpy ninja pyyaml mkl mkl-include setuptools cmake cffi typing_extensions future six requests dataclasses` -`conda install -c magma-cuda101` +`conda install -c pytorch magma-cuda101` change to a tmp dir @@ -52,11 +54,14 @@ build pytorch and install it `clear && TORCH_CUDA_ARCH_LIST="3.5 6.1" python setup.py install` run the `torch_test.py` to verify that CUDA is available and version: + `True` `1.7.0a0+57bffc3` ### Building TorchVision +change to a tmp dir + `git clone https://github.com/pytorch/vision.git` `cd vision/`